The recent advancements in text-to-image (T2I) and text-to-video (T2V) generation have significantly focused on enhancing safety, scalability, and user preference alignment. Researchers are increasingly adopting Direct Preference Optimization (DPO) strategies to refine generative models, ensuring they produce outputs that are not only high-quality but also aligned with user preferences and safety standards. Notably, the integration of low-rank adaptation (LoRA) matrices and novel merging strategies have enabled the removal of a broader range of harmful concepts from T2I models, showcasing a substantial leap in scalability and safety. In the realm of T2V generation, preference alignment has been advanced through the development of comprehensive preference scores and online optimization algorithms, addressing issues of visual quality, semantic alignment, and scalability. These innovations collectively push the boundaries of generative capabilities, offering more controlled and user-centric outputs. Notably, the use of large language models (LLMs) for prompt adaptation in video diffusion models has proven effective in generating preference-aligned prompts, further enhancing the precision and efficiency of video generation processes.